We extend latent Dirichlet allocation by introducing a topic model, hierarchically dual latent Dirichlet allocation, the output of which provides a basis for estimating consumers’ content preferences on the fly from their search queries. We extend latent Dirichlet allocation by introducing a topic model, hierarchically dual latent Dirichlet allocation (HDLDA), for contexts in which one type of document (e.g., search queries) are semantically related to another type of document (e.g., search results). In the context of online search engines, HDLDA identifies not only topics in short search queries and web pages, but also how the topics in search queries relate to the topics in the corresponding top search results. The output of HDLDA provides a basis for estimating consumers’ content preferences on the fly from their search queries given a set of assumptions on how consumers translate their content preferences into search queries. We apply HDLDA and explore its use in the estimation of content preferences in two studies. The first is a lab experiment in which we manipulate participants’ content preferences and observe the queries they formulate and their browsing behavior across different product categories. The second is a field study, which allows us to explore whether the content preferences estimated based on HDLDA may be used to explain and predict click-through rates in online search advertising.